It should be clear that documentation plays a major role in the design process that we are describing. Most programmers regard documentation as necessary evil, written as an afterthought only because some bureaucrat requires it. They do not expect it to be useful.
This is a self-fulfilling prophesy; documentation that has not been used before it is published, documentation that is not important to its author, will always be poor documentation.
Most of that documentation is incomplete and inaccurate, but those are not the main problems. If those were the main problems, the documents could be easily corrected by adding or correcting information. In fact, there are underlying organizational problems that lead to incompleteness and incorrectness and those problems, which are listed below, are not easily repaired.
1) Poor Organization: Most documentation today can be characterized as "stream of consciousness" and "stream of execution." "Stream of consciousness" writing puts information at the point in the text that the author was writing when the thought occurred to him. "Stream of execution" writing describes the system in the order things will happen when it runs. The problem with both of these documentation styles is that subsequent readers cannot find the information they seek. It will therefore not be easy to determine that facts are missing, or to correct them when they are wrong. It will not be easy to find all the parts of the document that should be changed when the software is changed. The documentation will be expensive to maintain and, in most cases, will not be maintained.
2) Boring Prose: Lots of words are used to say what could be said by a single programming language statement, a formula, or a diagram. Certain facts are repeated in many different sections. This increases the cost of the documentation and its maintenance. More importantly it leads to inattentive reading and undiscovered errors.
3) Confusing and Inconsistent Terminology: Any complex system requires the invention and definition of new terminology. Without it the documentation would be far too long. However, the writers of software documentation often fail to provide precise definitions for the terms they use. As a result, there are many terms used for the same concept and many similar but distinct concepts described with the same term.
4) Myopia: Documentation that is written when the project is nearing completion is written by people ho have lived with the system for so long that they take major decisions for granted. They document the small details that they think they will forget. Unfortunately, the result is a document useful to people who know the system well, but impenetrable for newcomers.
Documentation in the ideal design process meets the needs of the initial developers as well as the needs of the programmers who come later. Each of the documents mentioned above records requirements or design decisions and is used as a reference document for the rest of the design. However, they also provide the information that the maintainers will need. Because the documents are used as reference manuals throughout the building of the software, they will be mature and ready to use in later work. The documentation in this design process is not an afterthought; it is viewed as one of the primary products of the project. Some systematic checks can be applied to increase completeness and consistency. [...]
"Stream of consciousness" and "stream of execution" documentation is avoided by designing the structure of each document. Each document is designed by stating the questions that it must answer and refining the questions until each defines the content of an individual section. There must be one, and only one, place for every fact that will be in the document. The questions are answered, i.e., the document is written, only after the structure of the document has been defined. When there are several documents of a certain kind, a standard organization is written for those documents. Every document is designed in accordance with the same principle that guides our software design: separation of concerns. Each aspect of the system is described in exactly one section and nothing else is described in that section. When documents are reviewed, they are reviewed for adherence to the documentation rules as well as for accuracy.
The resulting documentation is not easy or relaxing reading, but it is not boring. It makes use of tables, formulas, and other formal notation to increase the density of information. The organizational rules prevent the duplication of information. The result is documentation that must be read very attentively, but rewards its reader with detailed and precise information. [...]
No matter how often we stumble on our way, the final documentation will be rational and accurate. Even mathematics, the discipline that many of us regard as the most rational of all follows this procedure. [...] Analogous reasoning applies to software. Those who read the software documentation want to understand the programs, not relive their discovery. By presenting rationalized documentation we provide what they need.
Our documentation differs from the ideal documentation in one important way. We make a policy of recording all of the design alternatives that we considered and rejected. For each, we explain why it was considered and why it was finally rejected. Months, weeks, or even hours later, when we wonder why we did what we did, we can find out. Years from now, the maintainer will have many of the same questions and will find his answers in our documents.
For much of my life, I have been a software voyeur, peeking furtively at other people's dirty code. Occasionally, I find a real jewel, a well-structured program written in a consistent style, free of kludges, developed so that each component is simple and organized, and designed so that the product is easy to change. Why, since we have been studying software construction for more than 30 years, don't we find more such jewels? How often is it possible to produce such a jewel of a system? Seldom? Frequently? Always?
Often, software has grown large and its structure degraded because designers have repeatedly modified it to integrate new systems or add new features. ... Wirth suggests that we keep our software lean by sticking to essentials, omitting "bells and whistles". Besides, lean software is likely to be smaller and even faster.
However, it isn't always necessary to choose between function and elegance. Perhaps, I'm too optimistic, but I don't think a designer must omit features to build what Wirth calls "lean software". What is necessary is to design the product so that newly added features do not eliminate useful capabilities, make good use of capabilities already present for other purposes, and can be ignored or deleted by people who don't want them. ... Given a choice between tool and jewel, we will choose tool; but with a little more thought, we can often have both. Studying the jewels can show us how.
One of the weaknesses of technological society is that we sometimes place far too much emphasis on originality. Creativity and originality are obviously valuable wherever there is room for improvement, and they are essential when dealing with problems for which we have no adequate solution.
Nevertheless, we have an unfortunate tendency to value creativity as an end in itself and use it as an excuse for ignorance. I have known both researchers and developers who refused to look at previous work because they wanted to use their own ideas. Managers often do not allow their designers time to study the way things have been done in the past. It seems obvious that we should use our own ideas only if they are better than previous ones. Successful innovators usually know previous work and have managed to understand the fundamental weakness in earlier approaches. Too many software products show evidence of "ignorant originality." They make the same mistakes others made before them and ignore solutions that others have found.
Sometimes new languages are used in the design of jewels, and authors may attribute a product's success to the use of a particular language or type of language. Here, I have grave doubts. ... My experience does not support the view that the programming language used determines the quality of the software. I have seen beautiful, lean software written only using assembler (Dijkstra offers an example), good software written in Fortran, and even good software written in C. I have also seen programs in which each of these tools was used badly. ... Focusing on the programming language is a red herring that will distract us from real solutions to the problem of poor software. The jewels I've found owe their elegance to: the use of good decomposition principles, the use of good hierarchical structures, and the design of interfaces.
There is much to learn from jewel-like systems. ... The most important lesson is "up-front investment." In each of the jewels I've seen, the designers had obviously spent a lot of time thinking about the structure of the system before writing code. The system structure could be accurately described and documented without reference to the code. Programs were not just written; they had been planned, often in some pseudocode or a language other than the actual programming language. ... My engineering teachers laid down some basic rules:
These rules apply to software at least as much as they do to circuits or machines.
The purpose of professional programming is to deliver a product that satisfies its users. The primary means of doing so is to produce software with a clean internal structure and to grow a group of designers and programmers skilled enough and motivated enough to respond quickly and effectively to change and opportunities.
Why? The internal structure of the program and the process by which it was created are ideally of no concern to the end user. Stronger: if the end user has to worry about how the program was written, then there is something wrong with the program. Given that, what is the importance of the structure of a program and of the people who create the program? A program needs a clean internal structure to ease: testing, porting, maintenance, extension, reorganization, and understanding.
The main point is that every successful piece of software has an extended life in which it is worked on by a succession of programmers and designers, ported to new hardware, adapted to unanticipated uses, and repeatedly reorganized. Throughout the software's life, new versions of it must be produced with acceptable error rates and on time.
Documentation connotes the creation of an artifact: namely, a document, which may, of course, be electronic files, Web pages, or paper. Thus, documenting a software architecture becomes a concrete task: producing a software architecture document. Viewing the activity as creating a tangible product has advantages. We can describe good architecture documents and bad ones. We can use completeness criteria to judge how much work is left in producing this artifact and determining when the task is done. Planning or tracking a project's progress around the creation of artifacts, or documents, is an excellent way to manage. Making the architecture information available to its consumers and keeping it up-to-date reduces to a solved problem of configuration control. Documentation can be formal or not, as appropriate. Documents may describe, or they may specify. Hence, the term is appropriately general.
Finally, documentation is a longstanding software engineering tradition. Documentation is the task that you are supposed to do because it's good for you. It's what your software engineering teachers taught you to do, your customers contracted you to do, your managers nagged you to do, and what you always found a way not to do. So if documentation brings up too many pangs of professional guilt, use any term you like that's more palatable. The essence of the activity is writing down--and keeping current--the results of architectural decisions so that the stakeholders of the architecture--people who need to know what it is to do their job--have the information in accessible, nonambiguous form.
Perhaps the most important concept associated with software architecture documentation is that of the view. A software architecture is a complex entity that cannot be described in a single one-dimensional fashion. Our analogy with the bird wing proves illuminating. There is no single rendition of a bird wing. Instead there are many: feathers, skeleton, circulation, muscular views, and many others. Which one of these views is the "architecture" of the wing? None of them. Which views convey the architecture? All of them.
We use the concept of views to give us the most fundamental principle of architecture documentation: Documenting an architecture is a matter of documenting the relevant views and then adding documentation that applies to more than one view.
What are relevant views? It depends on your goals. As we saw previously, architecture documentation can serve many purposes: a mission statement for implementers, a basis for analysis, the specification for automatic code generation, the starting point for system understanding and asset recovery, or the blueprint for project planning. The views you document depend on the uses you expect to make of the documentation.
It may be disconcerting that no single view can fully represent an architecture. Additionally, it feels somehow inadequate to see the system only through discrete, multiple views, that may or may not relate to one another in a straightforward way. The essence of architecture is the suppression of information not necessary to the task at hand, and so it is somehow fitting that the very nature of architecture is such that it never presents its whole self to us but only a facet or two at a time. This is its strength: Each view emphasizes certain aspects of the system while deemphasizing or ignoring other aspects, all in the interest of making the problem at hand tractable. Nevertheless, no one of these individual views adequately documents the software architecture for the system. That is accomplished by the complete set of views along with information that transcends them.
Seven rules for sound documentation: 1) Write documentation from the reader's point of view. 2) Avoid unnecessary repetition. 3) Avoid ambiguity. 4) Use a standard organization. 5) Record rationale. 6) Keep documentation current but not too current. 7) Review documentation for fitness of purpose.
A good program works flawlessly and has no bugs. But what internal qualities produce such perfection? It's no mystery, we just need some occasional reminding. Whether you code in C/C++, C#, Java, Basic, Perl, COBOL, ASM, all good programming exhibits the same time-honored qualities: simplicity, readability, modularity, layering, design, efficiency, elegance, and clarity.
Simplicity means you don't do in ten lines what you can do in five. It means you make extra effort to be concise, but not to the point of obfuscation. It means you abhor open coding and functions that span pages. Simplicity--of organization, implementation, design--makes your code more reliable and bug free. There's less to go wrong.
Readability means what it says: that others can read your code. Readability means you bother to write comments, to follow conventions, and pause to name your variables wisely.
Modularity means your program is built like the universe. The world is made of molecules, which are made of atoms, electrons, nucleons, and quarks. Likewise good programs erect large systems from smaller ones, which are built from even smaller building blocks. And just as atoms combine in novel ways, software components should be reusable.
Layering means that internally, your program resembles a layer cake. The app sits on the framework sits on the OS sits on the hardware. Even within your app, you need layers, like file-document-view-frame. Higher layers call ones below, which raise events back up. (Calls go down; events go up.) Lower layers should never know what higher ones are up to. The essence of an event/callback is to provide blind upward notification. If your doc calls the frame directly, something stinks. Modules and layers are defined by APIs, which delineate their boundaries. Thus, design is critical.
Design means you take the time to plan your program before you build it. Thoughts are cheaper than debugging. A good rule of thumb is to spend half your time on design. You need a functional spec (what the program does) and an internal blueprint. APIs should be codified in writing.
Efficiency means your program is fast and economical. It doesn't hog files, data connections, or anything else. It does what it should, but no more. It loads and departs without fuss. At the function level, you can always optimize later, during testing. But at high levels, you must plan for performance. If the design requires a million trips to the server, expect a dog.
Elegance is like beauty: hard to describe but easy to recognize. Elegance combines simplicity, efficiency, and brilliance, and produces a feeling of pride. Elegance is when you replace a procedure with a table, or realize that you can use recursion--which is almost always elegant.
Clarity is the granddaddy of good programming, the platinum quality all the others serve. Computers make it possible to create systems that are vastly more complex than physical machines. The fundamental challenge of programming is managing complexity. Simplicity, readability, modularity, layering, design, efficiency, and elegance are all time-honored ways to achieve clarity, which is the antidote to complexity.
Clarity of code. Clarity of design. Clarity of purpose. You must understand--really understand--what you're doing at every level. Otherwise you're lost. Bad programs are less often a failure of coding skill than of having a clear goal. That's why design is key. It keeps you honest. If you can't write it down, if you can't explain it to others, you don't really know what you're doing.